Dark frames

I've been quite impressed by the results of simply pointing my Olympus
E10 at the sky and opening the shutter for eight seconds; it's got a
good f/2.4 lens, and I get images down to about magnitude 8.5 through
dire suburban skies.

On the other hand, it seems to have hot pixels. I thought hot pixels
were essentially a *manufacturing* fault in the CCD, and so a single
dark frame (leave aperture and speed settings fixed, just also leave
the lens-cap on) would characterise them adequately for all time. But
in a set of 100 photos taken last Tuesday, I'm still seeing hot pixels
after subtracting a dark-frame taken two weeks ago.

My temptation is to compute a minimum for each pixel position across
all the photos and use that as a dark frame; the photometry is
dreadful in any case, even after I've summed across 2x2 pixel groups -
not sure if that's a matter of the CCD response and the V filter being
vastly different, I should probably plot brightness-residual against
spectral type and see if there's a correlation. Is this a sensible
way to proceed, given that most of the photos are star fields?

Should I in fact be taking a dark frame for each observing session,
and if so does this indicate there's something wrong with the camera?

Tom